3 ms·
TFA is a big simplistic but, indeed, cache does rule everything. When you need really fast algorithms and very low latencies, you have to get your hands dirty.
by martinced 14y ago
TFA is a big simplistic but, indeed, cache does rule everything.
When you need really fast algorithms and very low latencies, you have to get your hands dirty.
If you're interested in much more advanced writings on the subject I'd suggest googling for "How to do 100K+ TPS at less than 1ms latency".
It is explaining how the LMAX's disruptor pattern is now able to process 12 million events per second on a single core in... Java!
It explains, amongst other, the size and speed of the various caches and how to minimize cache misses.
- stbullard 14y agoFor the lazy: Slides: http://bit.ly/11aeJfc http://bit.ly/11aeJfc Video (InfoQ): http://www.infoq.com/presentations/LMAX http://www.infoq.com/presentations/LMAX Blogpost (2011): http://martinfowler.com/articles/lmax.html?t=1319912579 http://martinfowler.com/articles/lmax.html?t=1319912579 Discussion (2011): http://news.ycombinator.com/item?id=3173993 http://news.ycombinator.com/item?id=3173993
- brown9-2 14y agoThe "Disruptor" library mentioned can also be found at http://lmax-exchange.github.com/disruptor/ http://lmax-exchange.github.com/disruptor/. Funny that this was posted here as their whitepaper has been sitting in my Downloads folder for months waiting to be read. The slides mention "Hotspot likes small compact methods". Does anyone happen to know a source or have more information on this?